across all 3 measurement periods. Again, all three models estimated with good fit. Subsequently,
a model fit comparison test was performed to determine the model with the best fit for the
structural analysis.
A chi-square difference test was used to compare the most constrained invariance models
(Measurement-1) with each subsequent less constrained models (Measurement-2 and
Measurement-3) for each physical, social, and emotional measurement models. Measurement
models with invariant factor loadings only (Measurement-2) demonstrated a chi-square
difference that was significantly better than the most constrained models (Measurement-1) but
not significantly different from the 'no invariance models' (Measurement-3). Thus, the invariant
factor loadings only measurement models (Measurement-2) were selected as the best fitting
measurement models for the subsequent structural equation model estimation. Model goodness
of fit statistics are presented in Tables 4 20.
Structural Equation Models
Estimation of baseline (Time 1) structural equation model. The structural analysis
was performed using the following procedures. First, a simple mediating, one occasion structural
model (Step 1), was specified for each pain outcome construct. The purpose of this initial, simple
(one-occasion) model was to determine whether the derived measurement model fits the
structural equation model. Each model was structured with direct paths from the exogenous
variables (age, sex, race, and number of diagnosed diseases) to pain, a direct path from pain to
PMI, and a direct path from PMI to each of the pain outcome latent constructs, conducted
separately for physical functioning, social functioning, and emotional functioning. The fit of
these models were poor, and using the chi-square comparison test, these models estimated
significantly worse than the single occasion Time 1 measurement model. Further model building
was required. See model fit indices in Table 4-21 and diagram in Figure 4-8.
67